security/device_cgroup.c

   1 /*
   2  * device_cgroup.c - device cgroup subsystem
   3  *
   4  * Copyright 2007 IBM Corp
   5  */
   6
   7 #include <linux/device_cgroup.h>
   8 #include <linux/cgroup.h>
   9 #include <linux/ctype.h>
  10 #include <linux/list.h>
  11 #include <linux/uaccess.h>
  12 #include <linux/seq_file.h>
  13 #include <linux/slab.h>
  14 #include <linux/rcupdate.h>
  15 #include <linux/mutex.h>
  16
  17 #define ACC_MKNOD 1
  18 #define ACC_READ  2
  19 #define ACC_WRITE 4
  20 #define ACC_MASK (ACC_MKNOD | ACC_READ | ACC_WRITE)
  21
  22 #define DEV_BLOCK 1
  23 #define DEV_CHAR  2
  24 #define DEV_ALL   4  /* this represents all devices */
  25
  26 static DEFINE_MUTEX(devcgroup_mutex);
  27
  28 /*
  29  * whitelist locking rules:
  30  * hold devcgroup_mutex for update/read.
  31  * hold rcu_read_lock() for read.
  32  */
  33
  34 struct dev_whitelist_item {
  35         u32 major, minor;
  36         short type;
  37         short access;
  38         struct list_head list;
  39         struct rcu_head rcu;
  40 };
  41
  42 struct dev_cgroup {
  43         struct cgroup_subsys_state css;
  44         struct list_head whitelist;
  45         bool deny_all;
  46 };
  47
  48 static inline struct dev_cgroup *css_to_devcgroup(struct cgroup_subsys_state *s)
  49 {
  50         return container_of(s, struct dev_cgroup, css);
  51 }
  52
  53 static inline struct dev_cgroup *cgroup_to_devcgroup(struct cgroup *cgroup)
  54 {
  55         return css_to_devcgroup(cgroup_subsys_state(cgroup, devices_subsys_id));
  56 }
  57
  58 static inline struct dev_cgroup *task_devcgroup(struct task_struct *task)
  59 {
  60         return css_to_devcgroup(task_subsys_state(task, devices_subsys_id));
  61 }
  62
  63 struct cgroup_subsys devices_subsys;
  64
  65 static int devcgroup_can_attach(struct cgroup *new_cgrp,
  66                                 struct cgroup_taskset *set)
  67 {
  68         struct task_struct *task = cgroup_taskset_first(set);
  69
  70         if (current != task && !capable(CAP_SYS_ADMIN))
  71                 return -EPERM;
  72         return 0;
  73 }
  74
  75 /*
  76  * called under devcgroup_mutex
  77  */
  78 static int dev_whitelist_copy(struct list_head *dest, struct list_head *orig)
  79 {
  80         struct dev_whitelist_item *wh, *tmp, *new;
  81
  82         list_for_each_entry(wh, orig, list) {
  83                 new = kmemdup(wh, sizeof(*wh), GFP_KERNEL);
  84                 if (!new)
  85                         goto free_and_exit;
  86                 list_add_tail(&new->list, dest);
  87         }
  88
  89         return 0;
  90
  91 free_and_exit:
  92         list_for_each_entry_safe(wh, tmp, dest, list) {
  93                 list_del(&wh->list);
  94                 kfree(wh);
  95         }
  96         return -ENOMEM;
  97 }
  98
  99 /* Stupid prototype - don't bother combining existing entries */
 100 /*
 101  * called under devcgroup_mutex
 102  */
 103 static int dev_whitelist_add(struct dev_cgroup *dev_cgroup,
 104                         struct dev_whitelist_item *wh)
 105 {
 106         struct dev_whitelist_item *whcopy, *walk;
 107
 108         whcopy = kmemdup(wh, sizeof(*wh), GFP_KERNEL);
 109         if (!whcopy)
 110                 return -ENOMEM;
 111
 112         list_for_each_entry(walk, &dev_cgroup->whitelist, list) {
 113                 if (walk->type != wh->type)
 114                         continue;
 115                 if (walk->major != wh->major)
 116                         continue;
 117                 if (walk->minor != wh->minor)
 118                         continue;
 119
 120                 walk->access |= wh->access;
 121                 kfree(whcopy);
 122                 whcopy = NULL;
 123         }
 124
 125         if (whcopy != NULL)
 126                 list_add_tail_rcu(&whcopy->list, &dev_cgroup->whitelist);
 127         return 0;
 128 }
 129
 130 /*
 131  * called under devcgroup_mutex
 132  */
 133 static void dev_whitelist_rm(struct dev_cgroup *dev_cgroup,
 134                         struct dev_whitelist_item *wh)
 135 {
 136         struct dev_whitelist_item *walk, *tmp;
 137
 138         list_for_each_entry_safe(walk, tmp, &dev_cgroup->whitelist, list) {
 139                 if (walk->type == DEV_ALL)
 140                         goto remove;
 141                 if (walk->type != wh->type)
 142                         continue;
 143                 if (walk->major != ~0 && walk->major != wh->major)
 144                         continue;
 145                 if (walk->minor != ~0 && walk->minor != wh->minor)
 146                         continue;
 147
 148 remove:
 149                 walk->access &= ~wh->access;
 150                 if (!walk->access) {
 151                         list_del_rcu(&walk->list);
 152                         kfree_rcu(walk, rcu);
 153                 }
 154         }
 155 }
 156
 157 /**
 158  * dev_whitelist_clean - frees all entries of the whitelist
 159  * @dev_cgroup: dev_cgroup with the whitelist to be cleaned
 160  *
 161  * called under devcgroup_mutex
 162  */
 163 static void dev_whitelist_clean(struct dev_cgroup *dev_cgroup)
 164 {
 165         struct dev_whitelist_item *wh, *tmp;
 166
 167         list_for_each_entry_safe(wh, tmp, &dev_cgroup->whitelist, list) {
 168                 list_del(&wh->list);
 169                 kfree(wh);
 170         }
 171 }
 172
 173 /*
 174  * called from kernel/cgroup.c with cgroup_lock() held.
 175  */
 176 static struct cgroup_subsys_state *devcgroup_create(struct cgroup *cgroup)
 177 {
 178         struct dev_cgroup *dev_cgroup, *parent_dev_cgroup;
 179         struct cgroup *parent_cgroup;
 180         int ret;
 181
 182         dev_cgroup = kzalloc(sizeof(*dev_cgroup), GFP_KERNEL);
 183         if (!dev_cgroup)
 184                 return ERR_PTR(-ENOMEM);
 185         INIT_LIST_HEAD(&dev_cgroup->whitelist);
 186         parent_cgroup = cgroup->parent;
 187
 188         if (parent_cgroup == NULL) {
 189                 struct dev_whitelist_item *wh;
 190                 wh = kmalloc(sizeof(*wh), GFP_KERNEL);
 191                 if (!wh) {
 192                         kfree(dev_cgroup);
 193                         return ERR_PTR(-ENOMEM);
 194                 }
 195                 wh->minor = wh->major = ~0;
 196                 wh->type = DEV_ALL;
 197                 wh->access = ACC_MASK;
 198                 dev_cgroup->deny_all = false;
 199                 list_add(&wh->list, &dev_cgroup->whitelist);
 200         } else {
 201                 parent_dev_cgroup = cgroup_to_devcgroup(parent_cgroup);
 202                 mutex_lock(&devcgroup_mutex);
 203                 ret = dev_whitelist_copy(&dev_cgroup->whitelist,
 204                                 &parent_dev_cgroup->whitelist);
 205                 dev_cgroup->deny_all = parent_dev_cgroup->deny_all;
 206                 mutex_unlock(&devcgroup_mutex);
 207                 if (ret) {
 208                         kfree(dev_cgroup);
 209                         return ERR_PTR(ret);
 210                 }
 211         }
 212
 213         return &dev_cgroup->css;
 214 }
 215
 216 static void devcgroup_destroy(struct cgroup *cgroup)
 217 {
 218         struct dev_cgroup *dev_cgroup;
 219
 220         dev_cgroup = cgroup_to_devcgroup(cgroup);
 221         dev_whitelist_clean(dev_cgroup);
 222         kfree(dev_cgroup);
 223 }
 224
 225 #define DEVCG_ALLOW 1
 226 #define DEVCG_DENY 2
 227 #define DEVCG_LIST 3
 228
 229 #define MAJMINLEN 13
 230 #define ACCLEN 4
 231
 232 static void set_access(char *acc, short access)
 233 {
 234         int idx = 0;
 235         memset(acc, 0, ACCLEN);
 236         if (access & ACC_READ)
 237                 acc[idx++] = 'r';
 238         if (access & ACC_WRITE)
 239                 acc[idx++] = 'w';
 240         if (access & ACC_MKNOD)
 241                 acc[idx++] = 'm';
 242 }
 243
 244 static char type_to_char(short type)
 245 {
 246         if (type == DEV_ALL)
 247                 return 'a';
 248         if (type == DEV_CHAR)
 249                 return 'c';
 250         if (type == DEV_BLOCK)
 251                 return 'b';
 252         return 'X';
 253 }
 254
 255 static void set_majmin(char *str, unsigned m)
 256 {
 257         if (m == ~0)
 258                 strcpy(str, "*");
 259         else
 260                 sprintf(str, "%u", m);
 261 }
 262
 263 static int devcgroup_seq_read(struct cgroup *cgroup, struct cftype *cft,
 264                                 struct seq_file *m)
 265 {
 266         struct dev_cgroup *devcgroup = cgroup_to_devcgroup(cgroup);
 267         struct dev_whitelist_item *wh;
 268         char maj[MAJMINLEN], min[MAJMINLEN], acc[ACCLEN];
 269
 270         rcu_read_lock();
 271         list_for_each_entry_rcu(wh, &devcgroup->whitelist, list) {
 272                 set_access(acc, wh->access);
 273                 set_majmin(maj, wh->major);
 274                 set_majmin(min, wh->minor);
 275                 seq_printf(m, "%c %s:%s %s\n", type_to_char(wh->type),
 276                            maj, min, acc);
 277         }
 278         rcu_read_unlock();
 279
 280         return 0;
 281 }
 282
 283 /*
 284  * may_access_whitelist:
 285  * does the access granted to dev_cgroup c contain the access
 286  * requested in whitelist item refwh.
 287  * return 1 if yes, 0 if no.
 288  * call with devcgroup_mutex held
 289  */
 290 static int may_access_whitelist(struct dev_cgroup *c,
 291                                        struct dev_whitelist_item *refwh)
 292 {
 293         struct dev_whitelist_item *whitem;
 294
 295         list_for_each_entry(whitem, &c->whitelist, list) {
 296                 if (whitem->type & DEV_ALL)
 297                         return 1;
 298                 if ((refwh->type & DEV_BLOCK) && !(whitem->type & DEV_BLOCK))
 299                         continue;
 300                 if ((refwh->type & DEV_CHAR) && !(whitem->type & DEV_CHAR))
 301                         continue;
 302                 if (whitem->major != ~0 && whitem->major != refwh->major)
 303                         continue;
 304                 if (whitem->minor != ~0 && whitem->minor != refwh->minor)
 305                         continue;
 306                 if (refwh->access & (~whitem->access))
 307                         continue;
 308                 return 1;
 309         }
 310         return 0;
 311 }
 312
 313 /*
 314  * parent_has_perm:
 315  * when adding a new allow rule to a device whitelist, the rule
 316  * must be allowed in the parent device
 317  */
 318 static int parent_has_perm(struct dev_cgroup *childcg,
 319                                   struct dev_whitelist_item *wh)
 320 {
 321         struct cgroup *pcg = childcg->css.cgroup->parent;
 322         struct dev_cgroup *parent;
 323
 324         if (!pcg)
 325                 return 1;
 326         parent = cgroup_to_devcgroup(pcg);
 327         return may_access_whitelist(parent, wh);
 328 }
 329
 330 /*
 331  * Modify the whitelist using allow/deny rules.
 332  * CAP_SYS_ADMIN is needed for this.  It's at least separate from CAP_MKNOD
 333  * so we can give a container CAP_MKNOD to let it create devices but not
 334  * modify the whitelist.
 335  * It seems likely we'll want to add a CAP_CONTAINER capability to allow
 336  * us to also grant CAP_SYS_ADMIN to containers without giving away the
 337  * device whitelist controls, but for now we'll stick with CAP_SYS_ADMIN
 338  *
 339  * Taking rules away is always allowed (given CAP_SYS_ADMIN).  Granting
 340  * new access is only allowed if you're in the top-level cgroup, or your
 341  * parent cgroup has the access you're asking for.
 342  */
 343 static int devcgroup_update_access(struct dev_cgroup *devcgroup,
 344                                    int filetype, const char *buffer)
 345 {
 346         const char *b;
 347         char *endp;
 348         int count;
 349         struct dev_whitelist_item wh;
 350
 351         if (!capable(CAP_SYS_ADMIN))
 352                 return -EPERM;
 353
 354         memset(&wh, 0, sizeof(wh));
 355         b = buffer;
 356
 357         switch (*b) {
 358         case 'a':
 359                 wh.type = DEV_ALL;
 360                 wh.access = ACC_MASK;
 361                 wh.major = ~0;
 362                 wh.minor = ~0;
 363                 goto handle;
 364         case 'b':
 365                 wh.type = DEV_BLOCK;
 366                 break;
 367         case 'c':
 368                 wh.type = DEV_CHAR;
 369                 break;
 370         default:
 371                 return -EINVAL;
 372         }
 373         b++;
 374         if (!isspace(*b))
 375                 return -EINVAL;
 376         b++;
 377         if (*b == '*') {
 378                 wh.major = ~0;
 379                 b++;
 380         } else if (isdigit(*b)) {
 381                 wh.major = simple_strtoul(b, &endp, 10);
 382                 b = endp;
 383         } else {
 384                 return -EINVAL;
 385         }
 386         if (*b != ':')
 387                 return -EINVAL;
 388         b++;
 389
 390         /* read minor */
 391         if (*b == '*') {
 392                 wh.minor = ~0;
 393                 b++;
 394         } else if (isdigit(*b)) {
 395                 wh.minor = simple_strtoul(b, &endp, 10);
 396                 b = endp;
 397         } else {
 398                 return -EINVAL;
 399         }
 400         if (!isspace(*b))
 401                 return -EINVAL;
 402         for (b++, count = 0; count < 3; count++, b++) {
 403                 switch (*b) {
 404                 case 'r':
 405                         wh.access |= ACC_READ;
 406                         break;
 407                 case 'w':
 408                         wh.access |= ACC_WRITE;
 409                         break;
 410                 case 'm':
 411                         wh.access |= ACC_MKNOD;
 412                         break;
 413                 case '\n':
 414                 case '\0':
 415                         count = 3;
 416                         break;
 417                 default:
 418                         return -EINVAL;
 419                 }
 420         }
 421
 422 handle:
 423         switch (filetype) {
 424         case DEVCG_ALLOW:
 425                 if (!parent_has_perm(devcgroup, &wh))
 426                         return -EPERM;
 427                 devcgroup->deny_all = false;
 428                 return dev_whitelist_add(devcgroup, &wh);
 429         case DEVCG_DENY:
 430                 dev_whitelist_rm(devcgroup, &wh);
 431                 devcgroup->deny_all = true;
 432                 break;
 433         default:
 434                 return -EINVAL;
 435         }
 436         return 0;
 437 }
 438
 439 static int devcgroup_access_write(struct cgroup *cgrp, struct cftype *cft,
 440                                   const char *buffer)
 441 {
 442         int retval;
 443
 444         mutex_lock(&devcgroup_mutex);
 445         retval = devcgroup_update_access(cgroup_to_devcgroup(cgrp),
 446                                          cft->private, buffer);
 447         mutex_unlock(&devcgroup_mutex);
 448         return retval;
 449 }
 450
 451 static struct cftype dev_cgroup_files[] = {
 452         {
 453                 .name = "allow",
 454                 .write_string  = devcgroup_access_write,
 455                 .private = DEVCG_ALLOW,
 456         },
 457         {
 458                 .name = "deny",
 459                 .write_string = devcgroup_access_write,
 460                 .private = DEVCG_DENY,
 461         },
 462         {
 463                 .name = "list",
 464                 .read_seq_string = devcgroup_seq_read,
 465                 .private = DEVCG_LIST,
 466         },
 467         { }     /* terminate */
 468 };
 469
 470 struct cgroup_subsys devices_subsys = {
 471         .name = "devices",
 472         .can_attach = devcgroup_can_attach,
 473         .create = devcgroup_create,
 474         .destroy = devcgroup_destroy,
 475         .subsys_id = devices_subsys_id,
 476         .base_cftypes = dev_cgroup_files,
 477 };
 478
 479 int __devcgroup_inode_permission(struct inode *inode, int mask)
 480 {
 481         struct dev_cgroup *dev_cgroup;
 482         struct dev_whitelist_item *wh;
 483
 484         rcu_read_lock();
 485
 486         dev_cgroup = task_devcgroup(current);
 487
 488         list_for_each_entry_rcu(wh, &dev_cgroup->whitelist, list) {
 489                 if (wh->type & DEV_ALL)
 490                         goto found;
 491                 if ((wh->type & DEV_BLOCK) && !S_ISBLK(inode->i_mode))
 492                         continue;
 493                 if ((wh->type & DEV_CHAR) && !S_ISCHR(inode->i_mode))
 494                         continue;
 495                 if (wh->major != ~0 && wh->major != imajor(inode))
 496                         continue;
 497                 if (wh->minor != ~0 && wh->minor != iminor(inode))
 498                         continue;
 499
 500                 if ((mask & MAY_WRITE) && !(wh->access & ACC_WRITE))
 501                         continue;
 502                 if ((mask & MAY_READ) && !(wh->access & ACC_READ))
 503                         continue;
 504 found:
 505                 rcu_read_unlock();
 506                 return 0;
 507         }
 508
 509         rcu_read_unlock();
 510
 511         return -EPERM;
 512 }
 513
 514 int devcgroup_inode_mknod(int mode, dev_t dev)
 515 {
 516         struct dev_cgroup *dev_cgroup;
 517         struct dev_whitelist_item *wh;
 518
 519         if (!S_ISBLK(mode) && !S_ISCHR(mode))
 520                 return 0;
 521
 522         rcu_read_lock();
 523
 524         dev_cgroup = task_devcgroup(current);
 525
 526         list_for_each_entry_rcu(wh, &dev_cgroup->whitelist, list) {
 527                 if (wh->type & DEV_ALL)
 528                         goto found;
 529                 if ((wh->type & DEV_BLOCK) && !S_ISBLK(mode))
 530                         continue;
 531                 if ((wh->type & DEV_CHAR) && !S_ISCHR(mode))
 532                         continue;
 533                 if (wh->major != ~0 && wh->major != MAJOR(dev))
 534                         continue;
 535                 if (wh->minor != ~0 && wh->minor != MINOR(dev))
 536                         continue;
 537
 538                 if (!(wh->access & ACC_MKNOD))
 539                         continue;
 540 found:
 541                 rcu_read_unlock();
 542                 return 0;
 543         }
 544
 545         rcu_read_unlock();
 546
 547         return -EPERM;
 548 }